1 repository with Claude Code or Codex plugins.
A benchmark for one AI coding agent failure class: blanket staging, disarmed git hooks, and destroyed work. Sealed blind holdouts, misses shown.
1 plugin · CASEY-OS/agent-guard-bench